SPH Basics

Introduction into
Smoothed Particle
Hydrodynamics

Ralf Klessen
Astrophysikalisches Institut Potsdam
Overview
• Some words about hydrodynamics
• SPH – The Standard Implementation
• Modifications and Extensions
• Some Caveats
Hydrodynamics:
• text books on hydrodynamics
◦ Landau & Lifshitz, Volume VI - Hydrodynamics
◦ Landau & Lifshitz, Volume X - Kinetic theory
◦ Reichl, A modern course in statistical physics, Wiley, New York (1998)
◦ Shu, The physics of astrophysics 2, Univ. Sc. Books, Mill Valley (1992)

• derivation
◦ gases and fluids are large ensembles of interacting particles
◦ −→ state of system is described by location in 6N dimensional
phase space f (N )(~q1...~qN , p~1...~pN )d~q1...d~qN d~p1...d~pN
◦ time evolution governed by ‘equation of motion’ for 6N -dim
probability distribution function f (N )
◦ f (N ) → f (n) by integrating over all but n coordinates −→
BBGKY hierarchy of equations of motion (after Born, Bogoli-
ubov, Green, Kirkwood and Yvon)
◦ physical observables are typically associated with 1- or 2-body
probability density f (1) or f (2)
◦ at lowest level of hierarchy: 1-body distribution function de-
scribes the probability of finding a particle at time t in the vol-
ume element d~q at ~q with momenta in the range d~p at p~.
◦ Boltzmann equation – equation of motion for f (1)
df ∂f ˙ ~
≡ + ~q · ∇q f + p~˙ · ∇
~p f
dt ∂t
∂f ~q f + F~ · ∇ ~p f = fc .
= + ~v · ∇
∂t
Hydrodynamics:
• derivation
◦ Boltzmann equation
df ∂f ˙ ~
≡ + ~q · ∇q f + p~˙ · ∇
~p f
dt ∂t
∂f ~q f + F~ · ∇ ~p f = fc .
= + ~v · ∇
∂t
→ first line: transformation from comoving to spatially fixed
coordinate system.
→ second line: velocity ~v = ~q˙ and force F~ = p~˙
→ all higher order terms are ’hidden’ in the collision term fc

◦ observable quantities are typically (velocity) moments of the

Boltzmann equation, e.g.
→ density: Z
ρ= m f (~q, p~, t)d~p
→ momentum: Z
ρ~v = m~v f (~q, p~, t)d~p
→ kinetic energy density:
2
Z
ρ~v = m~v 2 f (~q, p~, t)d~p
Hydrodynamics:
• derivation
◦ in general: the i-th velocity moment hξii (of ξi = m~v i) is
1Z
hξii = ξi f (~q, p~, t)d~p
n
with the mean particle number density n defined as
Z
n= f (~q, p~, t) d~p

◦ the equation of motion for hξii is

∂f
 
~ ~ ~
Z   Z
ξi  + ~v · ∇q f + F · ∇p f  d~p = ξi {fc} d~p ,
∂t
which after some complicated rearrangement becomes

∂ ~ ~ ~
Z
nhξii + ∇q (nhξi~v i) + nF h∇p ξii = ξifc d~p
∂t
(Maxwell-Boltzmann transport equation for hξii)

◦ if the RHS is zero, then ξi is a conserved quantity. This is only

the case for first three moments, mass ξ0 = m, momentum
ξ~1 = m~v , and kinetic energy ξ2 = m~v 2/2.
◦ MB equations build a hierarically nested set of equations, as hξii
~q (nhξi~v i) and because the collision term
depends on hξi+1i via ∇
cannot be reduced to depend on ξi only.
−→ need for a closure equation
−→ in hydrodynamics this is typically the equation of state.
Hydrodynamics:
• assumptions
◦ continuum limit:
→ distribution function f must be a ‘smoothly’ varying function
on the scales of interest −→ local average possible
→ stated differently: the averaging scale (i.e. scale of interest)
must be larger than the mean free path of individual particles
→ stated differently: microscopic behavior of particles can be
neglected
→ concept of fluid element must be meaningful
◦ only ‘short range forces’:
→ forces between particles are short range or saturate −→ col-
lective effects can be neglected
→ stated differently: correlation length of particles in the system
is finite (and smaller than the scales of interest)
• limitations
◦ shocks (scales of interest become smaller than mean free path)
◦ phase transitions (correlation length may become infinite)
◦ description of self-gravitating systems
◦ description of fully fractal systems
Hydrodynamics:
• the equations of hydrodynamics
◦ hydrodynamics ≡ book keeping problem
One must keep track of the ‘change’ of a fluid element due to
various physical processes acting on it. How do its ‘properties’
evolve under the influence of compression, heat sources, cooling,
etc.?
◦ Eulerian vs. Lagrangian point of view

Eulerian Lagrangian

consider spatially fixed volume element following motion of fluid element

◦ hydrodynamic equations = set of equations for the five con-

served quantities (ρ, ρ~v , ρ~v 2/2) plus closure equation (plus trans-
port equations for ‘external’ forces if present, e.g. gravity, mag-
netic field, heat sources, etc.)
Hydrodynamics:
• the equations of hydrodynamics
◦ equations of hydrodynamics

dρ ∂ρ ~ = −ρ∇
~ · ~v
= + ~v · ∇ρ (continuity equation)
dt ∂t
d~v ∂~v ~ 1~ ~ ~ 2 η! ~ ~
= + (~v · ∇)~v = − ∇p − ∇φ + η ∇ ~v + ζ + ∇(∇ · ~v )
dt ∂t ρ 3
(Navier-Stokes equation)
d ∂ ~ = T ds − p ∇
~ · ~v
= + ~v · ∇ (energy equation)
dt ∂t dt ρ
~ 2φ = 4πGρ
∇ (Poisson’s equation)

p = RρT (equation of state)

~2
B 1 ~ ~ ~
~ ~
FB = −∇ + (B · ∇)B (magnetic force)
8π 4π
∂B~
~ × (~v × B)
=∇ ~ (Lorentz equation)
∂t

ρ = density, ~v = velocity, p = pressure, φ = gravitational poten-

tial, ζ and η viscosity coefficients,  = ρ~v 2/2 = kinetic energy
density, T = temperature, s = entropy, R = gas constant, B ~ =
magnetic field (cgs units)
Hydrodynamics:
• the equations of hydrodynamics
◦ mass transport – continuity equation
dρ ∂ρ ~ = −ρ∇
~ · ~v
= + ~v · ∇ρ
dt ∂t
(conservation of mass)

◦ transport equation for momentum – Navier Stokes equation

d~v ∂~v 1
~ v = − ∇p− ~ ~ ~ ~v + ζ +
2 η !
~ ∇·~
~ v)
= +(~v · ∇)~ ∇φ+η ∇ ∇(
dt ∂t ρ 3
momentum change due to
→ pressure gradients: (−ρ−1 ∇p) ~
→ (self) gravity: −∇φ ~
→ viscous forces (internal friction, contains div(∂vi/∂xj ) terms):
~ 2~v + ζ + η ∇( ~ ∇~ · ~v )
 
η∇ 3
(conservation of momentum, general form of momentum trans-
port: ∂t(ρvi) = −∂j Πij )
◦ transport equation for internal energy
d ∂ ~ = T ds − p ∇ ~ · ~v
= + ~v · ∇
dt ∂t dt ρ
→ follows from the thermodynamic relation d = T ds−p dV =
T ds + p/ρ2dρ which described changes in  due to entropy
changed and to volume changes (compression, expansion)
→ for adiabatic gas the first term vanishes (s =constant)
→ heating sources, cooling processes can be incorporated in ds
(conservation of energy)
Hydrodynamics:
• the equations of hydrodynamics
◦ closure equation – equation of state
→ general form of equation of state p = p(T, ρ, ...)
→ ideal gas: p = RρT
→ special case – isothermal gas: p = c2sT (as RT = c2s)
SPH Basics:
• literature
◦ Benz, W., SPH, in ‘The Numerical Modeling of Nonlinear Stellar Pulsa-
tions’ ed. J. R. Buchler, Kluwer (1990)
◦ Monaghan, J. J., Particle Methods for Hydrodynamics, Comp. Phys. Re-
ports (1985)
◦ Monaghan, J. J., SPH, ARA&A (1992)

• concept of SPH
◦ ‘invented’ independently by Lucy (1977) and Gingold & Mon-
aghan (1977)
◦ originally proposed as Monte Carlo approach to calculate the
time evolution of gaseous systems
◦ more intuitively understood as interpolation scheme:
The fluid is represented by an ensemble of particles i, each
carrying mass mi, momentum mi~vi, and hydrodynamic
properties (like pressure pi, temperature Ti, internal energy
i, entropy si, etc.). The time evolution is governed by
the equation of motion plus additional equations to modify
the hydrodynamic properties of the particles. Hydrodynamic
observables are obtained by a local averaging process.
SPH Basics:
• properties of local averaging processes
◦ local averages hf (~r)i for any quantity f (~r) can be obtained by
convolution with an appropriate smoothing function W (~r, ~h):
f (~r 0)W (~r − ~r 0, ~h) d3r0 .
Z
hf (~r)i ≡

the function W (~r, ~h) is called smoothing kernel

◦ the kernel must satisfy the following two conditions:
W (~r, ~h) d3r = 1 hf (~r)i −→ f (~r) for ~h → 0
Z
and
the kernel W therefore follows the same definitions as Dirac’s
delta function δ(~r): limh→0 W (~r, h) = δ(~r).
◦ most SPH implementations use spherical kernel functions
W (~r, ~h) ≡ W (r, h) with r = |~r| and h = |~h|.
(one could also use triaxial kernels, e.g. Martel et al. 1995)
◦ as the kernel function W can be seen as approximation to the
δ-function for small but finite h we can expand the averaged
function hf (~r)i into a Taylor series for h to obtain an estimate
for f (~r); if W is an even function, the first order term vanishes
and the errors are second order in h
hf (~r)i = f (~r) + O(h2)
this holds for functions f that are smooth and do not exhibit
steep gradients over the size of W (→ problems in shocks).
~ 2 f (~r) + O(h3 ))
(more specifically the expansion is hf (~r)i = f (~r) + κh2 ∇
SPH Basics:
• properties of local averaging processes
◦ within its intrinsic accuracy, the smoothing process therefore is
a linear function with respect to summation and multiplication:
hf (~r) + g(~r)i = hf (~r)i + hg(~r)i
hf (~r) · g(~r)i = hf (~r)i · hg(~r)i
(one follows from the linearity of integration with respect to summation, and two is true to O(h2 ))

◦ derivatives can be ‘drawn into’ the averaging process:

d *
d +
hf (~r)i = f (~r)
dt dt
~ (~r)i = h∇f
∇hf ~ (~r)i

Furthermore, the spatial derivative of f can be transformed into

a spatial derivative of W (no need for finite differences or grid):
~ ~ ~ (|~r − ~r 0|, h) d3r0 .
 
∇ hf (~r )i = ∇f (~r ) = f (~r 0) ∇W
Z

(shown by integrating by parts and assuming that the surface term vanishes; if the solution space is

extended far enough, either the function f itself or the kernel approach zero)

◦ basic concept of SPH is a particle representation of the fluid

−→ integration transforms into summation over discrete set of
particles; example density ρ:
hρ(~ri)i = mj W (|~ri − ~rj |, h) .
X

in this picture, the mass of each particle is smeared out over

its kernel region; the density at each location is obtained by
summing over the contributions of the various particles −→
smoothed particle hydrodynamics!
SPH Basics:
• properties of local averaging processes
◦ ‘scatter’ versus ‘gather’ approach:
hρ(~ri)i = mj W (|~ri − ~rj |, h) .
X

allows for two different interpretations...

1. particle i collects the contributions from all other particles j
which smoothing volumes hj scatter onto location ~ri
h → hj , i.e. use W (|~ri − ~rj |, hj ) in the summation
2. particle i gathers the contributions from all particles which
centers fall within the smoothing volume of i
h → hi, i.e. use W (|~ri − ~rj |, hi) in the summation

if all particles have the same smoothing length h = hi = hj

both approaches are equivalent; otherwise different j contribute
to the sum −→ violation of Newton’s 3. law!!
therefore, enforce force anti-symmetry by using the (arithmetic)
average of the smoothing lengths for all particle pairs
hi + hj
h −→ hij = .
2
SPH Basics:
• the kernel function
◦ different functions meet the requirement W (|~r|, h) d3r = 1
R

and limh→0 W (|~r − ~r 0|, h)f (~r 0) d3r0 = f (~r):

R

→ Gaussian kernel:
2
 
1 r 
W (r, h) = 3/2 3 exp  −
h2

π h
· pro: mathematically sound
· pro: derivatives exist to all orders and are smooth
· contra: all particles contribute to a location
→ spline functions with compact support
→ the standard kernel: cubic spline
with ξ = r/h it is defined as


 1 − 32 ξ 2 + 34 ξ 3, for 0 ≤ ξ ≤ 1;
1



1
(2 − ξ)3,

W (r, h) ≡ 3 4 for 1 ≤ ξ ≤ 2;
πh  

 0,


otherwise.
· pro: compact support −→ all interactions are zero for
r > 2h −→ number of particles involved in the average
remains small (typically between 30 and 80)
· pro: second derivative is continuous
· pro: dominant error term is second order in h
→ in principle different kernel functions could be used for dif-
ferent equations (but it brings no obvious advantage, except
maybe in the case of XSPH)
→ specialized kernels can be constructed for different types of
problems
SPH Basics:
• variable smoothing length h
◦ spatial resolution of SPH is limited by h, the scale over which
forces and physical properties are smeared out
◦ to make optimum use of the Lagrangian nature of SPH one has
to allow for variations of h: in high-density regions h should be
small, in regions of low density h should be large
◦ the optimum value of h is such that every particle has ∼ 50
neighbors within the smoothing volume
◦ caveats:
→ introduction of additional errors (the Taylor series now con-
~
tains contributions from ∇h, furthermore time derivatives
∂h/∂t occur); however, these errors are of second or higher
order and thus the same as the one inherent to SPH anyway
→ modification of the kernel gradient
~ 0 ~ 0
 ∂ 0 ~

∇W (|~r−~r |, h) = ∇W (|~r−~r |, h)h+ W (|~r−~r |, h)∇h~r
 

∂h
the new term is ∝ ∇h ~ and becomes important only if the
smoothing length varies on scales less than the smoothing
lengths itself −→ it is generally neglected (see Nelson &
Papaloizou 1994).
◦ equation of ‘motion’ for h can be coupled to the density: from
h = h0 (ρ0/ρ)3 it follows from using the continuity equation
dh 1 h dρ 1 ~
=− = h ∇·~v (1)
dt 3 ρ dt 3
alternative methods exist (see e.g. Steinmetz & Müller 1993)
SPH Basics:
• the fluid equations in SPH
◦ there is an infinite number of possible SPH implementations of
the hydrodynamic equations!
◦ some notation: hij = (hi +hj )/2, ~rij = ~ri −~rj , ~vij = ~vi −~vj , and
~ i is the gradient with respect to the coordinates of particle i;
∇
all measurements are taken at particle positions (e.g. ρi = ρ(~ri))
◦ general form of SPH equations:

Ni mj
hfii = fj W (rij , hij )
X

j=1 ρj

◦ density — continuity equation (conservation of mass)

Ni
ρi = mj W (rij , hij )
X

j=1

dρi Ni
or =
X
~ iW (rij , hij )
mj~vij · ∇
dt j=1
(the second implementation is almost never used, see however
Monaghan 1991 for an application to water waves)
important
density is needed for ALL particles BEFORE computing
other averaged quantities −→ at each timestep, SPH com-
putations consist of TWO loops, first the density is obtained
for each particle, and then in a second round, all other par-
ticle properties are updated.

◦ pressure is defined via the equation of state (for example for

isothermal gas pi = c2sρi)
SPH Basics:
• the fluid equations in SPH
◦ velocity — Navier Stokes equation (conservation of momentum)
→ consider for now only pressure contributions: Euler’s equa-
tion
d~v ∂~v 1 ~  p  − p ∇ρ
 

= + (~v · ∇)~ ~ = −∇
~ v = − ∇p ~ (∗)
dt ∂t ρ ρ ρ 2

~
here, the identity ∇(pρ −1
) = ρ−1∇p ~ is used
~ − pρ−2∇ρ
→ in the SPH formalism this reads as
 
d~vi Ni pi pj  ~
=− mj  +  ∇iW (rij , hij )
X

ρ2i ρ2j

dt j=1

where the first term in (*) is neglected because it leads to

surface terms in the averaging procedure; it is assumed that
either the pressure or the kernel becomes zero at the inte-
gration border; if this is not the case correction terms need
to be added above.
the equation is anti-symmetric in i and j and conserves mo-
mentum locally and globally.
→ if self-gravity is taken into account, the gravitational force
needs to be added on the RHS
~ i = −G X mj rij
N
F~G = −∇φ 2
j=1 rij rij

note that the sum needs to be taken over ALL particles ←−

computationally expensive
SPH Basics:
• the fluid equations in SPH
◦ velocity — Navier Stokes equation (conservation of momentum)
→ the contribution of viscosity:
· converts ordered kinetic energy into random kinetic energy (heat)
· molecular viscosity in most astrophysical problems is small (except
maybe in shocks) −→ SPH normally has NO explicit treatment of
physical viscosity.
· however, artificial viscosity is needed to prevent particle interpene-
tration.
· this is achieved by smearing out shocks and by introducing dissi-
pation in regions with strong velocity divergence
· there are MANY ways to formulate artificial viscosity!!
→ the standard formulation of viscous pressures is
~ · ~v ) ,
pα = Πα ρ2 = −αρ`cs(∇
and
~ · ~v )2 .
pβ = Πβ ρ2 = −βρ`2(∇
α and β are free parameters and control the strength of the
viscous terms (typical values are α = 1 and β = 2); ` is the
scale over which shocks are smeared out (typically ` ≈ 2h).
· pα is a combined shear and bulk viscosity – it dampens
post-shock oscillations
· pβ is a von Neumann-Richtmyer viscosity – necessary to
prevent interpenetration in high Mach number shocks
SPH Basics:
• the fluid equations in SPH
◦ velocity — Navier Stokes equation (conservation of momentum)
→ the SPH implementation of the standard artificial viscosity is
Ni
F~ivisc = − ~ iW (rij , hij ) ,
mj Πij ∇
X

j=1
where the viscosity tensor Πij is defined by
(−αcij µij + βµ2ij )/ρij for ~vij · ~rij ≤ 0 ,



Πij =  0 for ~vij · ~rij > 0 ,
where
h~vij · ~rij
µij = 2 .
~rij + 0.01h2
with ~rij = ~ri − ~rj , ~vij = ~vi − ~vj , mean density ρij =
(ρi + ρj )/2, and mean sound speed cij = (ci + cj )/2.

→ Advantages of the standard artificial viscosity

· Galilean invariant
· vanishes for rigid body rotation (but not for differential rotation!!!)
· conserved linear and angular momenta
→ Disadvantages of the standard formula
· generates entropy in shear flows −→ Balsara viscosity
· leads to strong dissipation (one simulates ‘honey’ instead of inter-
stellar gas) −→ time-dependent viscosity & XSPH
· arbitrariness (no physical motivation) −→ Flebbe-type viscosities
→ many alternative formulations exist
→ set together, the momentum equation is
 
d~vi Ni pi pj
=−
X
mj  ∇
+ 2 + Πij  ~ iW (rij , hij ) − ∇φi
2

dt j=1 ρi ρj
SPH Basics:
• the fluid equations in SPH
◦ energy equation (conservation of momentum)
→ recall the hydrodynamic energy equation:
d ∂ ~ = ds − p ∇
~ · ~v
= + ~v · ∇
dt ∂t dt ρ
→ for adiabatic systems (c = const) the SPH form follows as
di pi XNi
= 2 ~ iW (rij , hij ) ,
mj ~vij · ∇
dt ρi j=1
(note that the alternative form
 
di 1X Ni pi pj
= ~ i W (rij , hij )
mj  2 + 2  ~vij · ∇
dt 2 j=1 ρi ρj

can lead to unphysical solutions, like negative internal energy)

→ dissipation due to (artificial) viscosity leads to a term
di 1 X Ni
= ~ iW (rij .hij )
mj Πij ~vij · ∇
dt 2 j=1

→ the presence of heating sources or cooling processes can be

incorporated into a function Γi.
→ altogether:
di pi XNi 1 XNi
= 2 ~
mj ~vij · ∇iWij + ~ iWij + Γi
mj Πij ~vij · ∇
dt ρi j=1 2 j=1
SPH Basics:
• the fluid equations in SPH
◦ entropy equation
→ alternatively to the energy equation one can integrate an
equation for the ‘entropy’
→ the entropic function A(s) is defined by
p = A(s)ργ .
the internal energy follows as
A(s) γ−1
= ρ (∗)
γ−1
→ the time evolution of A(s) depends on the emissivity per unit
volume Γ (heat sources and sinks) and on the viscosity; one
possible SPH implementation is
dAi γ−1 1γ − 1 XNi
=− Γi + ~ iW (rij , hij ) .
mj Πij~vij · ∇
γ−1
dt ρi 2 ρi =1

→ the time evolution of i is then derived from this equation via

(*), the temperature Ti of particle i is directly proportional
to i.
SPH Basics:
• time integration
◦ time integration is done similar to N -body methods
◦ there are two main schemes: leap-frog and predictor-corrector
methods
◦ variable timesteps
→ efficient use of CPU power in strongly inhomogeneous sys-
tems
→ typically, the lengths of timestep bins differ by factor 2
→ criteria for chosing the timestep
· Courant-Friedrichs-Lewy plus viscosity criterion
0.3 h
δtcv = .
~ ~
cs + h|∇ · ~v | + 1.2(αcs + βh|∇ · ~v |)
· force criterion v
h
u
u
δtf = 0.3 ,
u
u
t
~
|F |
· global error tolerance criteria are possible in Runge-Kutta
schemes
• boundary conditions
◦ closed (or periodic) boundaries can be handled by introducing
‘ghost’ particles
◦ open boundaries are difficult, because of large pressure gradients
(e.g. water surface on air)
Modifications of SPH:
o

• alternative ways to force anti-symmetry

◦ instead of using one kernel and take a mean value for h, average
of the kernel contributions of each particle:
h + h 1
  ( )
i j
W |~ri − ~rj |, → W (|~ri − ~rj |, hi) + W (|~ri − ~rj |, hj )
2 2

◦ instead of the artithmetic mean for the quantity p/ρ2 use the
geometric one:
√
pi pj
 
1  pi pj 
+ −→
2 ρ2i ρ2j
 
ρi ρj

• XSPH (Monaghan 1989)

◦ in the standard formulation the particle is advanced by integrat-
ing
d~ri
= ~vi
dt
◦ it may be more physical (and in the spirit of SPH) when moving
the particle with the smoothed flow velocity
d~ri ˆ Ni m
ˆ j
= ~v i with ~v i = ~vi + η (~vi − ~vj )W (|~ri − ~rj |, hij )
X

dt j=1 ρij

where η ≈ 0.5.
◦ this allows for a strongly reduced artificial viscosity term −→
reach higher Reynolds numbers when modeling interstellar tur-
bulence
◦ XSPH also allows for the introduction of the Cassama-Holm
subgrid model of turbulence (Monaghan 2002 – astro-ph/0204118)
Modifications of SPH:
• alternative formulations of viscosity
◦ Balsara viscosity:
→ standard viscosity generates entropy in shear flows (Balsara
1989)
→ add a correction term ∝ ∇~ × ~v
→ new viscosity:
(−αcij µij + βµ2ij )/ρij for ~vij · ~rij ≤ 0 ,



Πij = 
 0 for ~vij · ~rij > 0 ,
where now
h~vij · ~rij fi + fj
µij =
~rij2 + 0.01h2 2
with
~ · ~v |i
|∇
fi =
~ · ~v |i + |∇
|∇ ~ × ~v |i + 0.0001ci/h

→ this representation vanishes in pure shear flows, but is iden-

tical to the standard version in purely compressional flows

◦ for more physically motivated viscosity see

→ Flebbe et al., ApJ, 431, 754 (1994)
→ Watkins et al., ApJS, 119, 177 (1996)
→ etc
Modifications of SPH:
• alternative formulations of viscosity
◦ switch to reduce viscosity (Morris & Monaghan 1997)
→ artificial viscosity is a strongly undesired quantity, as it leads
to dissipation that is much higher than in astrophysical gases
→ for realistic models one wants as little artificial viscosity as
possible (e.g. important for turbulence simulations — we
model ‘honey’ instead of interstellar gas)
→ in priciple, artificial viscosity is only needed in regions of
strong compression (shocks)
→ introduce a switch which leads to high Πij when ∇ ~ · ~v be-
comes strongly negative and then let Πij ‘decay’ to zero af-
terwards
→ implementation:
· each particle i carries its own value αi (and βi, e.g. with
βi = 2αi)
· time evolution
αi = αmin + A exp(−t/τ )
with decay time τ ≈ 10h/cs.
Modifications of SPH:
• fully conservative formulation using Lagrange
multipliers
- Springel & Hernquist (2002, astro-ph/0111016)
- Monaghan (2002, astro-ph/0204118)
◦ the Lagrangian for compressible flows which are generated by
the thermal energy (ρ, s) acts as effective potential is
1
 

ρ  v 2 − u(ρ, s) d3r.

Z  
L=
2
equations of motion follow with s = const from
d ∂L ∂L
− =0
dt ∂~v ∂~r
◦ after some SPH arithmetics, one can derive the following accel-
eration equation for particle i
 
d~vi Ni  1 pi 1 pj ~
~
 

=− mj  ∇ W (r , h ) + ∇ W (r , h )
X
 f ρ2
i ij i i ij j 
dt j=1 i i fj ρ2j 

where
hi ∂ρi 
 

fi = 1 +
3ρi ∂hi

◦ the Lagrange multiplier used here is the constraint that hi is ad-

justed such that each smoothing volume contains a fixed amount
of mass
◦ under this contraint, the formulation conserves energy, entropy,
linear and angular momentum (∇h~ terms are taken into account
implicitely)
Properties of SPH:
• no clear mathematical convergence study −→ reliability of method
needs to rely on comparison with analytic solutions and on empirical
tests (e.g. comparing results obtained with different particle num-
bers)
• SPH is more dissipative than most grid-based methods
• SPH is Lagrangian, it can resolve large density contrasts whereever
needed (regions of interest need not to be defined in advance)
• SPH provides good resolution in high-density regions, however, only
poorly resolves low-density regions
• SPH generally performes poorly when handling shocks (but see
GPH)
• SPH is a particle scheme −→ good for describing the transition
from gaseous to stellar-dynamical systems (i.e. good for describing
the formation of stellar clusters)
• SPH cannot (yet) handle magnetic field satisfactory (problems with
~ ·B
stability and with ∇ ~ = 0 requirement)

• SPH can be combined with the special purpose hardware GRAPE